The Ministry of the Interior and Safety announced on the 10th that it will apply an AI deepfake detection model, developed jointly with the National Forensic Service, to the ninth nationwide local elections this June.
The move follows a surge in deepfake crimes that precisely synthesize a person’s face or voice. Officials say the measure is intended to prevent false material that manipulates a candidate’s image or remarks from distorting voters’ judgments during the election process.
The detection model was built on results from the “AI Detection Model Competition to Counter Deepfake Crimes,” held in December 2025.
The competition drew 268 teams (1,077 participants). The ministry and the National Forensic Service selected five top-performing models to provide to the National Election Commission and will operate a support system to rapidly analyze suspicious content during the election period.
The model improves detection accuracy by combining “global analysis,” which examines a video’s overall flow, with “local analysis,” which detects manipulation traces in specific areas such as the face.
It was designed to handle content generated by the latest generative AI. In validation tests, the model achieved about 92% detection accuracy.
The ministry and the National Forensic Service plan to expand the model’s use in phases and to establish a tightly coordinated, whole-of-government digital-crime response system with the Ministry of Gender Equality and Family, the Broadcasting, Media and Communications Commission, the National Police Agency, and other agencies.
Interior Minister Yun Ho-jung said, “Fake news using AI and deepfakes created with advanced technology seriously threaten electoral democracy. The government will steadily strengthen its capacity to counter false and manipulated information during elections to protect citizens’ fundamental rights and to build a capable, trusted ‘AI democratic government.’”
